"""Wrapper around MemoryDB vector database."""
from __future__ import annotations
import logging
import os
import uuid
from typing import (
TYPE_CHECKING,
Any,
Callable,
Dict,
Iterable,
List,
Mapping,
Optional,
Tuple,
Type,
Union,
cast,
)
import numpy as np
import yaml # type: ignore[import-untyped]
from langchain_core._api import deprecated
from langchain_core.callbacks import (
AsyncCallbackManagerForRetrieverRun,
CallbackManagerForRetrieverRun,
)
from langchain_core.documents import Document
from langchain_core.embeddings import Embeddings
from langchain_core.vectorstores import VectorStore, VectorStoreRetriever
from langchain_aws.utilities.redis import (
_array_to_buffer,
_buffer_to_array,
get_client,
)
from langchain_aws.utilities.utils import maximal_marginal_relevance
from langchain_aws.vectorstores.inmemorydb.constants import (
INMEMORYDB_TAG_SEPARATOR,
)
logger = logging.getLogger(__name__)
ListOfDict = List[Dict[str, str]]
if TYPE_CHECKING:
from redis.client import Redis as InMemoryDBType # type: ignore[import-untyped]
from redis.commands.search.query import Query # type: ignore[import-untyped]
from langchain_aws.vectorstores.inmemorydb.filters import InMemoryDBFilterExpression
from langchain_aws.vectorstores.inmemorydb.schema import InMemoryDBModel
def _default_relevance_score(val: float) -> float:
return 1 - val
[docs]def check_index_exists(client: InMemoryDBType, index_name: str) -> bool:
"""Check if MemoryDB index exists."""
try:
client.ft(index_name).info()
except: # noqa: E722
logger.debug("Index does not exist")
return False
logger.debug("Index already exists")
return True
[docs]class InMemoryVectorStore(VectorStore):
"""InMemoryVectorStore vector database.
To use, you should have the ``redis`` python package installed
for AWS MemoryDB
.. code-block:: bash
Once running, you can connect to the MemoryDB server with the following url schemas:
- redis://<host>:<port> # simple connection
- redis://<username>:<password>@<host>:<port> # connection with authentication
- rediss://<host>:<port> # connection with SSL
- rediss://<username>:<password>@<host>:<port> # connection with SSL and auth
Examples:
The following examples show various ways to use the Redis VectorStore with
LangChain.
For all the following examples assume we have the following imports:
.. code-block:: python
from langchain_aws.vectorstores import InMemoryVectorStore
Initialize, create index, and load Documents
.. code-block:: python
from langchain_aws.vectorstores import InMemoryVectorStore
rds = InMemoryVectorStore.from_documents(
documents, # a list of Document objects from loaders or created
embeddings, # an Embeddings object
redis_url="redis://cluster_endpoint:6379",
)
Initialize, create index, and load Documents with metadata
.. code-block:: python
rds = InMemoryVectorStore.from_texts(
texts, # a list of strings
metadata, # a list of metadata dicts
embeddings, # an Embeddings object
redis_url="redis://cluster_endpoint:6379",
)
Initialize, create index, and load Documents with metadata and return keys
.. code-block:: python
rds, keys = InMemoryVectorStore.from_texts_return_keys(
texts, # a list of strings
metadata, # a list of metadata dicts
embeddings, # an Embeddings object
redis_url="redis://cluster_endpoint:6379",
)
For use cases where the index needs to stay alive, you can initialize
with an index name such that it's easier to reference later
.. code-block:: python
rds = InMemoryVectorStore.from_texts(
texts, # a list of strings
metadata, # a list of metadata dicts
embeddings, # an Embeddings object
index_name="my-index",
redis_url="redis://cluster_endpoint:6379",
)
Initialize and connect to an existing index (from above)
.. code-block:: python
# must pass in schema and key_prefix from another index
existing_rds = InMemoryVectorStore.from_existing_index(
embeddings, # an Embeddings object
index_name="my-index",
schema=rds.schema, # schema dumped from another index
key_prefix=rds.key_prefix, # key prefix from another index
redis_url="redis://cluster_endpoint:6379",
)
Advanced examples:
Custom vector schema can be supplied to change the way that
MemoryDB creates the underlying vector schema. This is useful
for production use cases where you want to optimize the
vector schema for your use case. ex. using HNSW instead of
FLAT (knn) which is the default
.. code-block:: python
vector_schema = {
"algorithm": "HNSW"
}
rds = InMemoryVectorStore.from_texts(
texts, # a list of strings
metadata, # a list of metadata dicts
embeddings, # an Embeddings object
vector_schema=vector_schema,
redis_url="redis://cluster_endpoint:6379",
)
Custom index schema can be supplied to change the way that the
metadata is indexed. This is useful for you would like to use the
hybrid querying (filtering) capability of MemoryDB.
By default, this implementation will automatically generate the index
schema according to the following rules:
- All strings are indexed as text fields
- All numbers are indexed as numeric fields
- All lists of strings are indexed as tag fields (joined by
langchain_aws.vectorstores.inmemorydb.constants.INMEMORYDB_TAG_SEPARATOR)
- All None values are not indexed but still stored in MemoryDB these are
not retrievable through the interface here, but the raw MemoryDB client
can be used to retrieve them.
- All other types are not indexed
To override these rules, you can pass in a custom index schema like the following
.. code-block:: yaml
tag:
- name: credit_score
text:
- name: user
- name: job
Typically, the ``credit_score`` field would be a text field since it's a string,
however, we can override this behavior by specifying the field type as shown with
the yaml config (can also be a dictionary) above and the code below.
.. code-block:: python
rds = InMemoryVectorStore.from_texts(
texts, # a list of strings
metadata, # a list of metadata dicts
embeddings, # an Embeddings object
index_schema="path/to/index_schema.yaml", # can also be a dictionary
redis_url="redis://cluster_endpoint:6379",
)
When connecting to an existing index where a custom schema has been applied, it's
important to pass in the same schema to the ``from_existing_index`` method.
Otherwise, the schema for newly added samples will be incorrect and metadata
will not be returned.
"""
DEFAULT_VECTOR_SCHEMA = {
"name": "content_vector",
"algorithm": "FLAT",
"dims": 1536,
"distance_metric": "COSINE",
"datatype": "FLOAT32",
}
[docs] def __init__(
self,
redis_url: str,
index_name: str,
embedding: Embeddings,
index_schema: Optional[Union[Dict[str, ListOfDict], str, os.PathLike]] = None,
vector_schema: Optional[Dict[str, Union[str, int]]] = None,
relevance_score_fn: Optional[Callable[[float], float]] = None,
key_prefix: Optional[str] = None,
**kwargs: Any,
):
"""Initialize MemoryDB vector store with necessary components."""
self._check_deprecated_kwargs(kwargs)
self.index_name = index_name
self._embeddings = embedding
try:
redis_client = get_client(redis_url=redis_url, **kwargs)
except ValueError as e:
raise ValueError(f"Redis failed to connect: {e}")
self.client = redis_client
self.relevance_score_fn = relevance_score_fn
self._schema = self._get_schema_with_defaults(index_schema, vector_schema)
self.key_prefix = key_prefix if key_prefix is not None else f"doc:{index_name}"
@property
def embeddings(self) -> Optional[Embeddings]:
"""Access the query embedding object if available."""
return self._embeddings
[docs] @classmethod
def from_texts_return_keys(
cls,
texts: List[str],
embedding: Embeddings,
metadatas: Optional[List[dict]] = None,
index_name: Optional[str] = None,
index_schema: Optional[Union[Dict[str, ListOfDict], str, os.PathLike]] = None,
vector_schema: Optional[Dict[str, Union[str, int]]] = None,
**kwargs: Any,
) -> Tuple[InMemoryVectorStore, List[str]]:
"""Create a InMemoryVectorStore vectorstore from raw documents.
This is a user-friendly interface that:
1. Embeds documents.
2. Creates a new InMemoryVectorStore index if it doesn't already exist
3. Adds the documents to the newly created InMemoryVectorStore index.
4. Returns the keys of the newly created documents once stored.
This method will generate schema based on the metadata passed in
if the `index_schema` is not defined. If the `index_schema` is defined,
it will compare against the generated schema and warn if there are
differences. If you are purposefully defining the schema for the
metadata, then you can ignore that warning.
To examine the schema options, initialize an instance of this class
and print out the schema using the `InMemoryVectorStore.schema`` property. This
will include the content and content_vector classes which are
always present in the langchain schema.
Example:
.. code-block:: python
from langchain_aws.vectorstores import InMemoryVectorStore
embeddings = OpenAIEmbeddings()
redis, keys = InMemoryVectorStore.from_texts_return_keys(
texts,
embeddings,
redis_url="redis://cluster_endpoint:6379"
)
Args:
texts (List[str]): List of texts to add to the vectorstore.
embedding (Embeddings): Embeddings to use for the vectorstore.
metadatas (Optional[List[dict]], optional): Optional list of metadata
dicts to add to the vectorstore. Defaults to None.
index_name (Optional[str], optional): Optional name of the index to
create or add to. Defaults to None.
index_schema (Optional[Union[Dict[str, ListOfDict], str, os.PathLike]],
optional):
Optional fields to index within the metadata. Overrides generated
schema. Defaults to None.
vector_schema (Optional[Dict[str, Union[str, int]]], optional): Optional
vector schema to use. Defaults to None.
**kwargs (Any): Additional keyword arguments to pass to the Redis client.
Returns:
Tuple[InMemoryVectorStore, List[str]]:
Tuple of the InMemoryVectorStore instance and the keys of
the newly created documents.
Raises:
ValueError: If the number of metadatas does not match the number of texts.
"""
try:
import redis # type: ignore[import-untyped] # noqa: F401
from langchain_aws.vectorstores.inmemorydb.schema import read_schema
except ImportError as e:
raise ImportError(
"Could not import redis python package. "
"Please install it with `pip install redis`."
) from e
redis_url = kwargs.get("redis_url", os.getenv("REDIS_URL"))
if "redis_url" in kwargs:
kwargs.pop("redis_url")
# flag to use generated schema
if "generate" in kwargs:
kwargs.pop("generate")
# see if the user specified keys
keys = None
if "keys" in kwargs:
keys = kwargs.pop("keys")
# Name of the search index if not given
if not index_name:
index_name = uuid.uuid4().hex
# type check for metadata
if metadatas:
if isinstance(metadatas, list) and len(metadatas) != len(texts): # type: ignore
raise ValueError("Number of metadatas must match number of texts")
if not (isinstance(metadatas, list) and isinstance(metadatas[0], dict)):
raise ValueError("Metadatas must be a list of dicts")
generated_schema = _generate_field_schema(metadatas[0])
if index_schema:
# read in the schema solely to compare to the generated schema
user_schema = read_schema(index_schema) # type: ignore
# the very rare case where a super user decides to pass the index
# schema and a document loader is used that has metadata which
# we need to map into fields.
if user_schema != generated_schema:
logger.warning(
"`index_schema` does not match generated metadata schema.\n"
+ "If you meant to manually override the schema, please "
+ "ignore this message.\n"
+ f"index_schema: {user_schema}\n"
+ f"generated_schema: {generated_schema}\n"
)
else:
# use the generated schema
index_schema = generated_schema
# Create instance
# init the class -- if MemoryDB is unavailable, will throw exception
instance = cls(
redis_url,
index_name,
embedding,
index_schema=index_schema,
vector_schema=vector_schema,
**kwargs,
)
# Add data to MemoryDB
keys = instance.add_texts(texts, metadatas, keys=keys)
return instance, keys
[docs] @classmethod
def from_texts(
cls: Type[InMemoryVectorStore],
texts: List[str],
embedding: Embeddings,
metadatas: Optional[List[dict]] = None,
index_name: Optional[str] = None,
index_schema: Optional[Union[Dict[str, ListOfDict], str, os.PathLike]] = None,
vector_schema: Optional[Dict[str, Union[str, int]]] = None,
**kwargs: Any,
) -> InMemoryVectorStore:
"""Create a InMemoryVectorStore vectorstore from a list of texts.
This is a user-friendly interface that:
1. Embeds documents.
2. Creates a new InMemoryVectorStore index if it doesn't already exist
3. Adds the documents to the newly created InMemoryVectorStore index.
This method will generate schema based on the metadata passed in
if the `index_schema` is not defined. If the `index_schema` is defined,
it will compare against the generated schema and warn if there are
differences. If you are purposefully defining the schema for the
metadata, then you can ignore that warning.
To examine the schema options, initialize an instance of this class
and print out the schema using the `InMemoryVectorStore.schema`` property. This
will include the content and content_vector classes which are
always present in the langchain schema.
Example:
.. code-block:: python
from langchain_aws.vectorstores import InMemoryVectorStore
embeddings = OpenAIEmbeddings()
Args:
texts (List[str]): List of texts to add to the vectorstore.
embedding (Embeddings): Embedding model class (i.e. OpenAIEmbeddings)
for embedding queries.
metadatas (Optional[List[dict]], optional): Optional list of metadata dicts
to add to the vectorstore. Defaults to None.
index_name (Optional[str], optional): Optional name of the index to create
or add to. Defaults to None.
index_schema (Optional[Union[Dict[str, ListOfDict], str, os.PathLike]],
optional):
Optional fields to index within the metadata. Overrides generated
schema. Defaults to None.
vector_schema (Optional[Dict[str, Union[str, int]]], optional): Optional
vector schema to use. Defaults to None.
**kwargs (Any): Additional keyword arguments to pass to the
InMemoryVectorStore client.
Returns:
InMemoryVectorStore: InMemoryVectorStore VectorStore instance.
Raises:
ValueError: If the number of metadatas does not match the number of texts.
ImportError: If the redis python package is not installed.
"""
instance, _ = cls.from_texts_return_keys(
texts,
embedding,
metadatas=metadatas,
index_name=index_name,
index_schema=index_schema,
vector_schema=vector_schema,
**kwargs,
)
return instance
[docs] @classmethod
def from_existing_index(
cls,
embedding: Embeddings,
index_name: str,
schema: Union[Dict[str, ListOfDict], str, os.PathLike, Dict[str, ListOfDict]],
key_prefix: Optional[str] = None,
**kwargs: Any,
) -> InMemoryVectorStore:
"""Connect to an existing InMemoryVectorStore index.
Example:
.. code-block:: python
from langchain_aws.vectorstores import InMemoryVectorStore
embeddings = OpenAIEmbeddings()
# must pass in schema and key_prefix from another index
existing_rds = InMemoryVectorStore.from_existing_index(
embeddings,
index_name="my-index",
schema=rds.schema, # schema dumped from another index
key_prefix=rds.key_prefix, # key prefix from another index
redis_url="redis://username:password@cluster_endpoint:6379",
)
Args:
embedding (Embeddings): Embedding model class (i.e. OpenAIEmbeddings)
for embedding queries.
index_name (str): Name of the index to connect to.
schema (Union[Dict[str, str], str, os.PathLike, Dict[str, ListOfDict]]):
Schema of the index and the vector schema. Can be a dict, or path to
yaml file.
key_prefix (Optional[str]): Prefix to use for all keys in
InMemoryVectorStore associated
with this index.
**kwargs (Any): Additional keyword arguments to pass to the Redis client.
Returns:
InMemoryVectorStore: InMemoryVectorStore VectorStore instance.
Raises:
ValueError: If the index does not exist.
ImportError: If the redis python package is not installed.
"""
redis_url = kwargs.get("redis_url", os.getenv("REDIS_URL"))
# We need to first remove redis_url from kwargs,
# otherwise passing it to Redis will result in an error.
if "redis_url" in kwargs:
kwargs.pop("redis_url")
# Create instance
# init the class -- if InMemoryVectorStore is unavailable, will throw exception
instance = cls(
redis_url,
index_name,
embedding,
index_schema=schema,
key_prefix=key_prefix,
**kwargs,
)
# Check for existence of the declared index
if not check_index_exists(instance.client, index_name):
# Will only raise if the running InMemoryVectorStore server does not
# have a record of this particular index
# have a record of this particular index
raise ValueError(
f"InMemoryVectorStore failed to connect: "
f"Index {index_name} does not exist."
)
return instance
@property
def schema(self) -> Dict[str, List[Any]]:
"""Return the schema of the index."""
return self._schema.as_dict()
[docs] def write_schema(self, path: Union[str, os.PathLike]) -> None:
"""Write the schema to a yaml file."""
with open(path, "w+") as f:
yaml.dump(self.schema, f)
[docs] @staticmethod
def delete(
ids: Optional[List[str]] = None,
**kwargs: Any,
) -> bool:
"""
Delete a InMemoryVectorStore entry.
Args:
ids: List of ids (keys in redis) to delete.
redis_url: Redis connection url. This should be passed in the kwargs
or set as an environment variable: redis_url.
Returns:
bool: Whether or not the deletions were successful.
Raises:
ValueError: If the redis python package is not installed.
ValueError: If the ids (keys in redis) are not provided
"""
redis_url = kwargs.get("redis_url", os.getenv("REDIS_URL"))
if ids is None:
raise ValueError("'ids' (keys)() were not provided.")
try:
import redis # noqa: F401
except ImportError:
raise ImportError(
"Could not import redis python package. "
"Please install it with `pip install redis`."
)
try:
# We need to first remove redis_url from kwargs,
# otherwise passing it to InMemoryVectorStore will result in an error.
if "redis_url" in kwargs:
kwargs.pop("redis_url")
client = get_client(redis_url=redis_url, **kwargs)
except ValueError as e:
raise ValueError(f"Your redis connected error: {e}")
# Check if index exists
try:
client.delete(*ids)
logger.info("Entries deleted")
return True
except: # noqa: E722
# ids does not exist
return False
[docs] @staticmethod
def drop_index(
index_name: str,
delete_documents: bool,
**kwargs: Any,
) -> bool:
"""
Drop a InMemoryVectorStore search index.
Args:
index_name (str): Name of the index to drop.
delete_documents (bool): Whether to drop the associated documents.
Returns:
bool: Whether or not the drop was successful.
"""
redis_url = kwargs.get("redis_url", os.getenv("REDIS_URL"))
try:
import redis # noqa: F401
except ImportError:
raise ImportError(
"Could not import redis python package. "
"Please install it with `pip install redis`."
)
try:
# We need to first remove redis_url from kwargs,
# otherwise passing it to InMemoryVectorStore will result in an error.
if "redis_url" in kwargs:
kwargs.pop("redis_url")
client = get_client(redis_url=redis_url, **kwargs)
except ValueError as e:
raise ValueError(f"Your redis connected error: {e}")
# Check if index exists
try:
client.ft(index_name).dropindex(delete_documents)
logger.info("Drop index")
return True
except: # noqa: E722
# Index not exist
return False
[docs] def add_texts(
self,
texts: Iterable[str],
metadatas: Optional[List[dict]] = None,
embeddings: Optional[List[List[float]]] = None,
batch_size: int = 1000,
clean_metadata: bool = True,
**kwargs: Any,
) -> List[str]:
"""Add more texts to the vectorstore.
Args:
texts (Iterable[str]): Iterable of strings/text to add to the vectorstore.
metadatas (Optional[List[dict]], optional): Optional list of metadatas.
Defaults to None.
embeddings (Optional[List[List[float]]], optional): Optional pre-generated
embeddings. Defaults to None.
keys (List[str]) or ids (List[str]): Identifiers of entries.
Defaults to None.
batch_size (int, optional): Batch size to use for writes. Defaults to 1000.
Returns:
List[str]: List of ids added to the vectorstore
"""
ids = []
# Get keys or ids from kwargs
# Other vectorstores use ids
keys_or_ids = kwargs.get("keys", kwargs.get("ids"))
# type check for metadata
if metadatas:
if isinstance(metadatas, list) and len(metadatas) != len(texts): # type: ignore
raise ValueError("Number of metadatas must match number of texts")
if not (isinstance(metadatas, list) and isinstance(metadatas[0], dict)):
raise ValueError("Metadatas must be a list of dicts")
embeddings = embeddings or self._embeddings.embed_documents(list(texts))
self._create_index_if_not_exist(dim=len(embeddings[0]))
# Write data to InMemoryVectorStore
pipeline = self.client.pipeline(transaction=False)
for i, text in enumerate(texts):
# Use provided values by default or fallback
key = keys_or_ids[i] if keys_or_ids else str(uuid.uuid4().hex)
if not key.startswith(self.key_prefix + ":"):
key = self.key_prefix + ":" + key
metadata = metadatas[i] if metadatas else {}
metadata = _prepare_metadata(metadata) if clean_metadata else metadata
pipeline.hset(
key,
mapping={
self._schema.content_key: text,
self._schema.content_vector_key: _array_to_buffer(
embeddings[i], self._schema.vector_dtype
),
**metadata,
},
)
ids.append(key)
# Write batch
if i % batch_size == 0:
pipeline.execute()
# Cleanup final batch
pipeline.execute()
return ids
[docs] def as_retriever(self, **kwargs: Any) -> InMemoryVectorStoreRetriever:
tags = kwargs.pop("tags", None) or []
tags.extend(self._get_retriever_tags())
return InMemoryVectorStoreRetriever(vectorstore=self, **kwargs, tags=tags)
[docs] @deprecated("0.0.1", alternative="similarity_search(distance_threshold=0.1)")
def similarity_search_limit_score(
self, query: str, k: int = 4, score_threshold: float = 0.2, **kwargs: Any
) -> List[Document]:
"""
Returns the most similar indexed documents to the query text within the
score_threshold range.
Deprecated: Use similarity_search with distance_threshold instead.
Args:
query (str): The query text for which to find similar documents.
k (int): The number of documents to return. Default is 4.
score_threshold (float): The minimum matching *distance* required
for a document to be considered a match. Defaults to 0.2.
Returns:
List[Document]: A list of documents that are most similar to the query text
including the match score for each document.
Note:
If there are no documents that satisfy the score_threshold value,
an empty list is returned.
"""
return self.similarity_search(
query, k=k, distance_threshold=score_threshold, **kwargs
)
[docs] def similarity_search_with_score(
self,
query: str,
k: int = 4,
filter: Optional[InMemoryDBFilterExpression] = None,
return_metadata: bool = True,
**kwargs: Any,
) -> List[Tuple[Document, float]]:
"""Run similarity search with **vector distance**.
The "scores" returned from this function are the raw vector
distances from the query vector. For similarity scores, use
``similarity_search_with_relevance_scores``.
Args:
query (str): The query text for which to find similar documents.
k (int): The number of documents to return. Default is 4.
filter (InMemoryDBFilterExpression, optional): Optional metadata filter.
Defaults to None.
return_metadata (bool, optional): Whether to return metadata.
Defaults to True.
Returns:
List[Tuple[Document, float]]: A list of documents that are
most similar to the query with the distance for each document.
"""
try:
import redis
except ImportError as e:
raise ImportError(
"Could not import redis python package. "
"Please install it with `pip install redis`."
) from e
if "score_threshold" in kwargs:
logger.warning(
"score_threshold is deprecated. Use distance_threshold instead."
+ "score_threshold should only be used in "
+ "similarity_search_with_relevance_scores."
+ "score_threshold will be removed in a future release.",
)
query_embedding = self._embeddings.embed_query(query)
redis_query, params_dict = self._prepare_query(
query_embedding,
k=k,
filter=filter,
with_metadata=return_metadata,
with_distance=True,
**kwargs,
)
# Perform vector search
# ignore type because redis-py is wrong about bytes
try:
results = self.client.ft(self.index_name).search(redis_query, params_dict) # type: ignore
except redis.exceptions.ResponseError as e:
# split error message and see if it starts with "Syntax"
if str(e).split(" ")[0] == "Syntax":
raise ValueError(
"Query failed with syntax error. "
+ "This is likely due to malformation of "
+ "filter, vector, or query argument"
) from e
raise e
# Prepare document results
docs_with_scores: List[Tuple[Document, float]] = []
for result in results.docs:
metadata = {}
if return_metadata:
metadata = {"id": result.id}
metadata.update(self._collect_metadata(result))
doc = Document(page_content=result.content, metadata=metadata)
distance = self._calculate_fp_distance(result.distance)
docs_with_scores.append((doc, distance))
return docs_with_scores
[docs] def similarity_search(
self,
query: str,
k: int = 4,
filter: Optional[InMemoryDBFilterExpression] = None,
return_metadata: bool = True,
distance_threshold: Optional[float] = None,
**kwargs: Any,
) -> List[Document]:
"""Run similarity search
Args:
query (str): The query text for which to find similar documents.
k (int): The number of documents to return. Default is 4.
filter (InMemoryDBFilterExpression, optional): Optional metadata filter.
Defaults to None.
return_metadata (bool, optional): Whether to return metadata.
Defaults to True.
distance_threshold (Optional[float], optional): Maximum vector distance
between selected documents and the query vector. Defaults to None.
Returns:
List[Document]: A list of documents that are most similar to the query
text.
"""
query_embedding = self._embeddings.embed_query(query)
return self.similarity_search_by_vector(
query_embedding,
k=k,
filter=filter,
return_metadata=return_metadata,
distance_threshold=distance_threshold,
**kwargs,
)
[docs] def similarity_search_by_vector(
self,
embedding: List[float],
k: int = 4,
filter: Optional[InMemoryDBFilterExpression] = None,
return_metadata: bool = True,
distance_threshold: Optional[float] = None,
**kwargs: Any,
) -> List[Document]:
"""Run similarity search between a query vector and the indexed vectors.
Args:
embedding (List[float]): The query vector for which to find similar
documents.
k (int): The number of documents to return. Default is 4.
filter (InMemoryDBFilterExpression, optional): Optional metadata filter.
Defaults to None.
return_metadata (bool, optional): Whether to return metadata.
Defaults to True.
distance_threshold (Optional[float], optional): Maximum vector distance
between selected documents and the query vector. Defaults to None.
Returns:
List[Document]: A list of documents that are most similar to the query
text.
"""
try:
import redis
except ImportError as e:
raise ImportError(
"Could not import redis python package. "
"Please install it with `pip install redis`."
) from e
if "score_threshold" in kwargs:
logger.warning(
"score_threshold is deprecated. Use distance_threshold instead."
+ "score_threshold should only be used in "
+ "similarity_search_with_relevance_scores."
+ "score_threshold will be removed in a future release.",
)
redis_query, params_dict = self._prepare_query(
embedding,
k=k,
filter=filter,
distance_threshold=distance_threshold,
with_metadata=return_metadata,
with_distance=False,
)
# Perform vector search
# ignore type because redis-py is wrong about bytes
try:
results = self.client.ft(self.index_name).search(redis_query, params_dict) # type: ignore
except redis.exceptions.ResponseError as e:
# split error message and see if it starts with "Syntax"
if str(e).split(" ")[0] == "Syntax":
raise ValueError(
"Query failed with syntax error. "
+ "This is likely due to malformation of "
+ "filter, vector, or query argument"
) from e
raise e
# Prepare document results
docs = []
for result in results.docs:
metadata = {}
if return_metadata:
metadata = {"id": result.id}
metadata.update(self._collect_metadata(result))
content_key = self._schema.content_key
docs.append(
Document(page_content=getattr(result, content_key), metadata=metadata)
)
return docs
[docs] def max_marginal_relevance_search(
self,
query: str,
k: int = 4,
fetch_k: int = 20,
lambda_mult: float = 0.5,
filter: Optional[InMemoryDBFilterExpression] = None,
return_metadata: bool = True,
distance_threshold: Optional[float] = None,
**kwargs: Any,
) -> List[Document]:
"""Return docs selected using the maximal marginal relevance.
Maximal marginal relevance optimizes for similarity to query AND diversity
among selected documents.
Args:
query (str): Text to look up documents similar to.
k (int): Number of Documents to return. Defaults to 4.
fetch_k (int): Number of Documents to fetch to pass to MMR algorithm.
lambda_mult (float): Number between 0 and 1 that determines the degree
of diversity among the results with 0 corresponding
to maximum diversity and 1 to minimum diversity.
Defaults to 0.5.
filter (InMemoryDBFilterExpression, optional): Optional metadata filter.
Defaults to None.
return_metadata (bool, optional): Whether to return metadata.
Defaults to True.
distance_threshold (Optional[float], optional): Maximum vector distance
between selected documents and the query vector. Defaults to None.
Returns:
List[Document]: A list of Documents selected by maximal marginal relevance.
"""
# Embed the query
query_embedding = self._embeddings.embed_query(query)
# Fetch the initial documents
prefetch_docs = self.similarity_search_by_vector(
query_embedding,
k=fetch_k,
filter=filter,
return_metadata=return_metadata,
distance_threshold=distance_threshold,
**kwargs,
)
prefetch_ids = [doc.metadata["id"] for doc in prefetch_docs]
# Get the embeddings for the fetched documents
prefetch_embeddings = [
_buffer_to_array(
cast(
bytes,
self.client.hget(prefetch_id, self._schema.content_vector_key),
),
dtype=self._schema.vector_dtype,
)
for prefetch_id in prefetch_ids
]
# Select documents using maximal marginal relevance
selected_indices = maximal_marginal_relevance(
np.array(query_embedding), prefetch_embeddings, lambda_mult=lambda_mult, k=k
)
selected_docs = [prefetch_docs[i] for i in selected_indices]
return selected_docs
def _collect_metadata(self, result: "Document") -> Dict[str, Any]:
"""Collect metadata from MemoryDB.
Method ensures that there isn't a mismatch between the metadata
and the index schema passed to this class by the user or generated
by this class.
Args:
result (Document): redis.commands.search.Document object returned
from Redis.
Returns:
Dict[str, Any]: Collected metadata.
"""
# new metadata dict as modified by this method
meta = {}
for key in self._schema.metadata_keys:
try:
meta[key] = getattr(result, key)
except AttributeError:
# warning about attribute missing
logger.warning(
f"Metadata key {key} not found in metadata. "
+ "Setting to None. \n"
+ "Metadata fields defined for this instance: "
+ f"{self._schema.metadata_keys}"
)
meta[key] = None
return meta
def _prepare_query(
self,
query_embedding: List[float],
k: int = 4,
filter: Optional[InMemoryDBFilterExpression] = None,
distance_threshold: Optional[float] = None,
with_metadata: bool = True,
with_distance: bool = False,
) -> Tuple["Query", Dict[str, Any]]:
# Creates Redis query
params_dict: Dict[str, Union[str, bytes, float]] = {
"vector": _array_to_buffer(query_embedding, self._schema.vector_dtype),
}
# prepare return fields including score
return_fields = [self._schema.content_key]
if with_distance:
return_fields.append("distance")
if with_metadata:
return_fields.extend(self._schema.metadata_keys)
if distance_threshold:
params_dict["distance_threshold"] = distance_threshold
return (
self._prepare_range_query(
k, filter=filter, return_fields=return_fields
),
params_dict,
)
return (
self._prepare_vector_query(k, filter=filter, return_fields=return_fields),
params_dict,
)
def _prepare_range_query(
self,
k: int,
filter: Optional[InMemoryDBFilterExpression] = None,
return_fields: Optional[List[str]] = None,
) -> "Query":
try:
from redis.commands.search.query import Query
except ImportError as e:
raise ImportError(
"Could not import redis python package. "
"Please install it with `pip install redis`."
) from e
return_fields = return_fields or []
vector_key = self._schema.content_vector_key
base_query = f"@{vector_key}:[VECTOR_RANGE $distance_threshold $vector]"
if filter:
base_query = str(filter) + " " + base_query
query_string = base_query + "=>{$yield_distance_as: distance}"
return (
Query(query_string)
.return_fields(*return_fields)
.sort_by("distance")
.paging(0, k)
.dialect(2)
)
def _prepare_vector_query(
self,
k: int,
filter: Optional[InMemoryDBFilterExpression] = None,
return_fields: Optional[List[str]] = None,
) -> "Query":
"""Prepare query for vector search.
Args:
k: Number of results to return.
filter: Optional metadata filter.
Returns:
query: Query object.
"""
try:
from redis.commands.search.query import Query
except ImportError as e:
raise ImportError(
"Could not import redis python package. "
"Please install it with `pip install redis`."
) from e
return_fields = return_fields or []
query_prefix = "*"
if filter:
query_prefix = f"{str(filter)}"
vector_key = self._schema.content_vector_key
base_query = f"({query_prefix})=>[KNN {k} @{vector_key} $vector AS distance]"
query = (
Query(base_query)
.return_fields(*return_fields)
.sort_by("distance")
.paging(0, k)
.dialect(2)
)
return query
def _get_schema_with_defaults(
self,
index_schema: Optional[Union[Dict[str, ListOfDict], str, os.PathLike]] = None,
vector_schema: Optional[Dict[str, Union[str, int]]] = None,
) -> "InMemoryDBModel":
# should only be called after init of Redis (so Import handled)
from langchain_aws.vectorstores.inmemorydb.schema import (
InMemoryDBModel,
read_schema,
)
schema = InMemoryDBModel()
# read in schema (yaml file or dict) and
# pass to the Pydantic validators
if index_schema:
schema_values = read_schema(index_schema) # type: ignore
schema = InMemoryDBModel(**schema_values)
# ensure user did not exclude the content field
# no modifications if content field found
schema.add_content_field()
# if no content_vector field, add vector field to schema
# this makes adding a vector field to the schema optional when
# the user just wants additional metadata
try:
# see if user overrode the content vector
schema.content_vector
# if user overrode the content vector, check if they
# also passed vector schema. This won't be used since
# the index schema overrode the content vector
if vector_schema:
logger.warning(
"`vector_schema` is ignored since content_vector is "
+ "overridden in `index_schema`."
)
# user did not override content vector
except ValueError:
# set default vector schema and update with user provided schema
# if the user provided any
vector_field = self.DEFAULT_VECTOR_SCHEMA.copy()
if vector_schema:
vector_field.update(vector_schema)
# add the vector field either way
schema.add_vector_field(vector_field)
return schema
def _create_index_if_not_exist(self, dim: int = 1536) -> None:
try:
from redis.commands.search.indexDefinition import ( # type: ignore
IndexDefinition,
IndexType,
)
except ImportError:
raise ImportError(
"Could not import redis python package. "
"Please install it with `pip install redis`."
)
# Set vector dimension
# can't obtain beforehand because we don't
# know which embedding model is being used.
self._schema.content_vector.dims = dim
# Check if index exists
if not check_index_exists(self.client, self.index_name):
# Create MemoryDB Index
self.client.ft(self.index_name).create_index(
fields=self._schema.get_fields(),
definition=IndexDefinition(
prefix=[self.key_prefix], index_type=IndexType.HASH
),
)
def _calculate_fp_distance(self, distance: str) -> float:
"""Calculate the distance based on the vector datatype
Two datatypes supported:
- FLOAT32
- FLOAT64
if it's FLOAT32, we need to round the distance to 4 decimal places
otherwise, round to 7 decimal places.
"""
if self._schema.content_vector.datatype == "FLOAT32":
return round(float(distance), 4)
return round(float(distance), 7)
def _check_deprecated_kwargs(self, kwargs: Mapping[str, Any]) -> None:
"""Check for deprecated kwargs."""
deprecated_kwargs = {
"redis_host": "redis_url",
"redis_port": "redis_url",
"redis_password": "redis_url",
"content_key": "index_schema",
"vector_key": "vector_schema",
"distance_metric": "vector_schema",
}
for key, value in kwargs.items():
if key in deprecated_kwargs:
raise ValueError(
f"Keyword argument '{key}' is deprecated. "
f"Please use '{deprecated_kwargs[key]}' instead."
)
def _select_relevance_score_fn(self) -> Callable[[float], float]:
if self.relevance_score_fn:
return self.relevance_score_fn
metric_map = {
"COSINE": self._cosine_relevance_score_fn,
"IP": self._max_inner_product_relevance_score_fn,
"L2": self._euclidean_relevance_score_fn,
}
try:
return metric_map[self._schema.content_vector.distance_metric]
except KeyError:
return _default_relevance_score
def _generate_field_schema(data: Dict[str, Any]) -> Dict[str, Any]:
"""
Generate a schema for the search index in Redis based on the input metadata.
Given a dictionary of metadata, this function categorizes each metadata
field into one of the three categories:
- text: The field contains textual data.
- numeric: The field contains numeric data (either integer or float).
- tag: The field contains list of tags (strings).
Args
data (Dict[str, Any]): A dictionary where keys are metadata field names
and values are the metadata values.
Returns:
Dict[str, Any]: A dictionary with three keys "text", "numeric", and "tag".
Each key maps to a list of fields that belong to that category.
Raises:
ValueError: If a metadata field cannot be categorized into any of
the three known types.
"""
result: Dict[str, Any] = {
"text": [],
"numeric": [],
"tag": [],
}
for key, value in data.items():
# Numeric fields
try:
int(value)
result["numeric"].append({"name": key})
continue
except (ValueError, TypeError):
pass
# None values are not indexed as of now
if value is None:
continue
# if it's a list of strings, we assume it's a tag
if isinstance(value, (list, tuple)):
if not value or isinstance(value[0], str):
result["tag"].append({"name": key})
else:
name = type(value[0]).__name__
raise ValueError(
f"List/tuple values should contain strings: '{key}': {name}"
)
continue
# Check if value is string before processing further
if isinstance(value, str):
result["text"].append({"name": key})
continue
# Unable to classify the field value
name = type(value).__name__
raise ValueError(
"Could not generate MemoryDB index field type mapping "
+ f"for metadata: '{key}': {name}"
)
return result
def _prepare_metadata(metadata: Dict[str, Any]) -> Dict[str, Any]:
"""
Prepare metadata for indexing in Redis by sanitizing its values.
- String, integer, and float values remain unchanged.
- None or empty values are replaced with empty strings.
- Lists/tuples of strings are joined into a single string with a comma separator.
Args:
metadata (Dict[str, Any]): A dictionary where keys are metadata
field names and values are the metadata values.
Returns:
Dict[str, Any]: A sanitized dictionary ready for indexing in Redis.
Raises:
ValueError: If any metadata value is not one of the known
types (string, int, float, or list of strings).
"""
def raise_error(key: str, value: Any) -> None:
raise ValueError(
f"Metadata value for key '{key}' must be a string, int, "
+ f"float, or list of strings. Got {type(value).__name__}"
)
clean_meta: Dict[str, Union[str, float, int]] = {}
for key, value in metadata.items():
if value is None:
clean_meta[key] = ""
continue
# No transformation needed
if isinstance(value, (str, int, float)):
clean_meta[key] = value
# if it's a list/tuple of strings, we join it
elif isinstance(value, (list, tuple)):
if not value or isinstance(value[0], str):
clean_meta[key] = INMEMORYDB_TAG_SEPARATOR.join(value)
else:
raise_error(key, value)
else:
raise_error(key, value)
return clean_meta
[docs]class InMemoryVectorStoreRetriever(VectorStoreRetriever):
"""Retriever for InMemoryVectorStore."""
vectorstore: InMemoryVectorStore
"""InMemoryVectorStore."""
search_type: str = "similarity"
"""Type of search to perform. Can be either
'similarity',
'similarity_distance_threshold',
'similarity_score_threshold'
"""
search_kwargs: Dict[str, Any] = {
"k": 4,
"score_threshold": 0.9,
# set to None to avoid distance used in score_threshold search
"distance_threshold": None,
}
"""Default search kwargs."""
allowed_search_types = [
"similarity",
"similarity_distance_threshold",
"similarity_score_threshold",
"mmr",
]
"""Allowed search types."""
class Config:
"""Configuration for this pydantic object."""
arbitrary_types_allowed = True
def _get_relevant_documents(
self, query: str, *, run_manager: CallbackManagerForRetrieverRun
) -> List[Document]:
if self.search_type == "similarity":
docs = self.vectorstore.similarity_search(query, **self.search_kwargs)
elif self.search_type == "similarity_distance_threshold":
if self.search_kwargs["distance_threshold"] is None:
raise ValueError(
"distance_threshold must be provided for "
+ "similarity_distance_threshold retriever"
)
docs = self.vectorstore.similarity_search(query, **self.search_kwargs)
elif self.search_type == "similarity_score_threshold":
docs_and_similarities = (
self.vectorstore.similarity_search_with_relevance_scores(
query, **self.search_kwargs
)
)
docs = [doc for doc, _ in docs_and_similarities]
elif self.search_type == "mmr":
docs = self.vectorstore.max_marginal_relevance_search(
query, **self.search_kwargs
)
else:
raise ValueError(f"search_type of {self.search_type} not allowed.")
return docs
async def _aget_relevant_documents(
self, query: str, *, run_manager: AsyncCallbackManagerForRetrieverRun
) -> List[Document]:
if self.search_type == "similarity":
docs = await self.vectorstore.asimilarity_search(
query, **self.search_kwargs
)
elif self.search_type == "similarity_distance_threshold":
if self.search_kwargs["distance_threshold"] is None:
raise ValueError(
"distance_threshold must be provided for "
+ "similarity_distance_threshold retriever"
)
docs = await self.vectorstore.asimilarity_search(
query, **self.search_kwargs
)
elif self.search_type == "similarity_score_threshold":
docs_and_similarities = (
await self.vectorstore.asimilarity_search_with_relevance_scores(
query, **self.search_kwargs
)
)
docs = [doc for doc, _ in docs_and_similarities]
elif self.search_type == "mmr":
docs = await self.vectorstore.amax_marginal_relevance_search(
query, **self.search_kwargs
)
else:
raise ValueError(f"search_type of {self.search_type} not allowed.")
return docs
[docs] def add_documents(self, documents: List[Document], **kwargs: Any) -> List[str]:
"""Add documents to vectorstore."""
return self.vectorstore.add_documents(documents, **kwargs)
[docs] async def aadd_documents(
self, documents: List[Document], **kwargs: Any
) -> List[str]:
"""Add documents to vectorstore."""
return await self.vectorstore.aadd_documents(documents, **kwargs)
